Policy learning, variable importance and causal prediction
Many patients in critical care are at high risk of acute kidney injury. Renal replacement therapy can be life-saving, but also puts patients at risk, besides imposing a high financial and logistical burden to critical care units. Motivated by this, we studied the question of when to start renal replacement therapy, in a close collaboration with intensive care clinicians and nephrologists (Morzywolek et al., 2022a,b). In this talk, I will reflect on our experience with regard to learning an optimal treatment policy based on the Ghent University Intensive Care database. I will next expand on methodological developments motivated by this. In particular, I will briefly discuss methods aimed at quantifying the relative importance of patient characteristics in explaining treatment affect heterogeneity (Hines, Diaz-Ordaz and Vansteelandt, 2022). Next, I will discuss more extensively the ongoing development of a causal decision support system to predict how individual patients will fare when treatment is initiated within the next 24 hours versus later. For this, I will propose a novel Neyman-orthogonal learner, called i-learner or imputation-learner, which improves upon other orthogonal learners (such as DR-learner (Kennedy, 2020)) by guaranteeing predictions within the range of the outcome data.
References:
Morzywołek, P., Steen, J., Vansteelandt, S., Decruyenaere, J., Sterckx, S., & Van Biesen, W. (2022a). Timing of dialysis in acute kidney injury using routinely collected data and dynamic treatment regimes. Critical Care, 26(1), 365.
Morzywołek, P., Steen, J., Van Biesen, W., Decruyenaere, J., & Vansteelandt, S. (2022b). On estimation and cross‐validation of dynamic treatment regimes with competing risks. Statistics in Medicine, 41(26), 5258-5275.
Hines, O., Diaz-Ordaz, K., & Vansteelandt, S. (2022). Variable importance measures for heterogeneous causal effects. arXiv preprint arXiv:2204.06030.
Kennedy, E. H. (2020). Towards optimal doubly robust estimation of heterogeneous causal effects. arXiv preprint arXiv:2004.14497.
: